Self-adaptive adjustment method and system for nanometer grinding process

Abstract

The invention provides a self-adaptive adjustment method and system for a nanometer grinding process, and the method comprises the steps: obtaining force signal data, acoustic emission signal data, temperature signal data and vibration signal data, forming the data into grinding data, carrying out the feature extraction of the grinding data, obtaining feature parameters, obtaining a grinding state through the combination of empirical wavelet transform, and carrying out the self-adaptive adjustment of the nanometer grinding process. The method comprises the following steps: identifying a grinding state through topological data analysis and a hidden Markov model, obtaining an identification result, obtaining energy flow data, obtaining a process entropy yield according to the energy flow data, carrying out performance evaluation through the identification result and the process entropy yield, obtaining an evaluation result, and optimizing process parameters through an alternating direction multiplier method. According to the technical scheme, optimized technological parameters are obtained, self-adaptive adjustment is carried out according to the optimized technological parameters, and through the technical scheme based on micro-state accurate recognition and technological parameter intelligent optimization, the intelligent requirement for self-adaptive control in the grinding process is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing, in particular to a self-adaptive adjustment method and system for a nano-grinding process. BACKGROUND

[0002] With the wide application of nano-grinding technology in the field of precision manufacturing, achieving high precision and high efficiency in the grinding process has become a key direction of technological development.

[0003] However, the existing nano-grinding process has many problems, and relies on preset parameters or simple feedback adjustment, which is difficult to cope with dynamic disturbances such as abrasive wear and interface state changes. Such limitations result in poor consistency of processing precision, low efficiency, short tool life, and inability to meet the needs of high-precision precision manufacturing.

[0004] Therefore, there is an urgent need for a technical solution based on accurate identification of microstate and intelligent optimization of process parameters to meet the intelligent requirements of self-adaptive regulation in the grinding process. SUMMARY

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a self-adaptive adjustment method for a nano-grinding process, which comprises: In the process of nano-grinding, force signal data, acoustic emission signal data, temperature signal data and vibration signal data are obtained and composed into grinding data; Feature extraction is performed on the grinding data to obtain feature parameters, and the grinding state is obtained by combining empirical wavelet transform; The grinding state is identified based on topological data analysis and hidden Markov model to obtain an identification result; Energy flow data is obtained, process entropy production rate is obtained based on the energy flow data, performance evaluation is performed based on the identification result and the process entropy production rate, and an evaluation result is obtained; Process parameters are optimized based on the alternating direction multiplier method to obtain optimized process parameters, and self-adaptive adjustment is performed based on the optimized process parameters.

[0006] As a further scheme of the present application, in the process of nano-grinding, force signal data, acoustic emission signal data, temperature signal data and vibration signal data are obtained and composed into grinding data, which comprises: A three-axis force sensor, an acoustic emission sensor, a miniature infrared temperature measurement array and a vibration acceleration sensor are configured on the grinding head; The normal force, tangential force and moment data in the grinding process are obtained based on the three-axis force sensor as force signal data; The acoustic emission raw signal in the grinding process is obtained based on the acoustic emission sensor as acoustic emission signal data; Obtaining temperature distribution data in the grinding process as temperature signal data based on a miniature infrared temperature measurement array; Obtaining vibration amplitude and vibration frequency of the grinding head in the grinding process as vibration signal data based on a vibration acceleration sensor; Synchronously aligning, denoising and preprocessing the force signal data, the acoustic emission signal data, the temperature signal data and the vibration signal data to obtain grinding data.

[0007] As a further scheme of the present application, feature extraction is performed on the grinding data to obtain feature parameters, and the grinding state is obtained in combination with empirical wavelet transform, including: Multi-scale decomposition is performed on the acoustic emission signal data in the grinding data based on empirical wavelet transform to obtain a grain cutting component, a friction vibration component and a structure resonance component; Feature extraction is performed on the grain cutting component, the friction vibration component and the structure resonance component to obtain an energy proportion of the grain cutting component, a main frequency offset of the friction vibration component and a peak amplitude of the structure resonance component; Feature extraction is performed on the force signal data in the grinding data to obtain a peak force, an average force and a force fluctuation coefficient; Feature extraction is performed on the temperature signal data in the grinding data to obtain an average temperature and a temperature gradient; Feature extraction is performed on the vibration signal data in the grinding data to obtain a root mean square value and a peak frequency; A mapping relationship between the feature parameters and the grinding state is established, and the current grinding state is obtained based on the mapping relationship.

[0008] As a further scheme of the present application, the grinding state is identified based on topological data analysis and a hidden Markov model to obtain an identification result, including: A high-dimensional feature vector is constructed based on the feature parameters, and the high-dimensional feature vector is processed by topological data analysis to obtain a process state category; The process state category is taken as a hidden state set of the hidden Markov model, and a feature parameter sequence is taken as an observation sequence, and the hidden Markov model is trained to obtain a state transition probability matrix and an observation probability matrix; The feature parameters extracted in real time are input into the hidden Markov model, the posterior probability of the current state and the current grinding state category are obtained based on a forward algorithm, the confidence degree of the grinding state category is obtained based on the posterior probability, and the grinding state category and the corresponding confidence degree are taken as the identification result; The process state category includes a high-efficiency cutting state, a stable grinding state and a critical wear state.

[0009] As a further scheme of the present application, the energy flow data is acquired, the process entropy production rate is acquired based on the energy flow data, the performance evaluation is performed based on the identification result and the process entropy production rate, and the evaluation result is acquired, including: The spindle motor power consumption and the pressure system power consumption are acquired, the material removal energy is acquired based on the abrasive grain cutting component energy of the acoustic emission signal data, and the spindle motor power consumption, the pressure system power consumption and the material removal energy are composed into the energy flow data; The spindle motor power consumption and the pressure system power consumption are summed to acquire the input power data, and the material removal energy is taken as the useful work data; The processing time of the nanometer grinding process is acquired, and the process entropy production rate is acquired based on the input power data, the useful work data and the processing time; The performance evaluation index system is set, and the performance evaluation is performed in combination with the identification result to acquire the evaluation result.

[0010] As a further scheme of the present application, the performance evaluation index system is set, and the performance evaluation is performed in combination with the identification result to acquire the evaluation result, including: The material removal amount is acquired, the material removal rate is acquired based on the material removal amount and the processing time, the material removal rate is taken as the processing efficiency index, the surface shape error of the processed material is acquired, the surface shape error is taken as the processing precision index, the process entropy production rate is taken as the process stability index, and the performance evaluation index system is set based on the processing efficiency index, the processing precision index and the process stability index; Based on the performance evaluation index system, the grinding state category in the identification result is scored to acquire the scoring result; The scoring threshold is set, and the scoring result is compared with the scoring threshold, if the scoring result is greater than or equal to the scoring threshold, it is determined that the performance meets the standard; If the scoring result is less than the scoring threshold, it is determined that the performance does not meet the standard.

[0011] As a further scheme of the present application, the process parameters are optimized based on the alternating direction multiplier method to acquire the optimized process parameters, and the adaptive adjustment is performed based on the optimized process parameters, including: The historical production data is acquired, the performance non-meeting state in the evaluation result is analyzed in combination with the historical production data to confirm the to-be-optimized process parameter set, and the to-be-optimized process parameter set includes the trajectory parameter and the processing parameter; The double-layer nested optimization target is constructed, including the inner-layer optimization target and the outer-layer optimization target, the inner-layer optimization target is to maximize the processing precision and efficiency based on the current trajectory parameter, and the outer-layer optimization target is to minimize the process entropy production rate based on the process state; The double-layer nested optimization target is decomposed into a plurality of sub-problems based on the alternating direction multiplier method, and the optimal solution of each sub-problem is solved through alternating iteration; A change amount threshold is set, and when the process parameter change amount of two adjacent iterations is less than the change amount threshold, convergence is determined, the optimal trajectory parameter and the optimal machining parameter are obtained, and the optimal process parameter is formed; The control instruction is generated based on the optimized process parameter, and the control instruction is sent to the grinding equipment, and the grinding equipment is adaptively adjusted based on the control instruction.

[0012] As a further scheme of the present application, the double-layer nested optimization target is decomposed into a plurality of sub-problems based on the alternating direction multiplier method, and the optimal solution of each sub-problem is solved by alternating iteration, comprising: In the solving process, the inner-layer optimization sub-problem satisfies the constraint condition that the trajectory coverage density is greater than or equal to the preset density threshold and the idle stroke proportion is less than the preset proportion threshold; The outer-layer optimization sub-problem satisfies the constraint condition that the process entropy production rate is less than the entropy production rate threshold.

[0013] In another aspect, the embodiment of the present application also provides a self-adaptive adjustment system for a nano-grinding process, comprising: An acquisition module is configured to acquire force signal data, acoustic emission signal data, temperature signal data and vibration signal data in a nano-grinding process, and form grinding data, and acquire energy flow data; An extraction module is configured to perform feature extraction on the grinding data, acquire feature parameters, and acquire a grinding state by using an empirical wavelet transform; An identification module is configured to identify the grinding state based on topological data analysis and a hidden Markov model, and acquire an identification result; An evaluation module is configured to acquire a process entropy production rate based on the energy flow data, perform performance evaluation based on the identification result and the process entropy production rate, and acquire an evaluation result; An optimization module is configured to optimize process parameters based on the alternating direction multiplier method, and acquire optimized process parameters; An adjustment module is configured to perform adaptive adjustment based on the optimized process parameters.

[0014] Compared with the prior art, the present application has the following beneficial effects: by regulating the entropy yield rate through the process, accurately monitoring abnormal conditions in the processing process, simultaneously reducing equipment maintenance costs and production interruption risks, and ensuring the continuous and stable operation of the production process; by using the alternating direction multiplier method, adaptive adjustment can be completed in a very short time, rapid response to various changes in the processing process can be achieved, the real-time control requirements in industrial production can be met, and the processing process can be ensured to be in the best state at all times; by optimizing the process parameters and the process, the material removal efficiency is improved, the overall processing time is effectively shortened, and the output per unit time is improved, and through this technical scheme based on accurate identification of the microstate and intelligent optimization of the process parameters, the intelligent requirements of adaptive regulation in the grinding process are realized. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a step flow chart of a nano-grinding process adaptive adjustment method of the present application; Figure 2 is a step flow chart of step S2 of acquiring the grinding state in the nano-grinding process adaptive adjustment method of the present application; Figure 3 is a schematic diagram of a nano-grinding process adaptive adjustment system of the present application. DETAILED DESCRIPTION

[0016] The present application will be specifically described below in conjunction with the drawings of the specification, Figure 1 is an execution flow schematic diagram of a nano-grinding process adaptive adjustment method provided by an embodiment of the present application, Figure 2 is a step flow chart of step S2 of acquiring the grinding state in the nano-grinding process adaptive adjustment method of the present application, and the nano-grinding process adaptive adjustment method will be described in detail below.

[0017] Step S1, in the nano-grinding process, force signal data, acoustic emission signal data, temperature signal data, and vibration signal data are acquired and composed into grinding data.

[0018] Further, a three-axis force sensor, an acoustic emission sensor, a miniature infrared temperature measurement array, and a vibration acceleration sensor are configured on the grinding head.

[0019] For example, the three-axis force sensor can be installed at the connection end of the grinding head and the main shaft, the acoustic emission sensor can be fixed on the side wall of the grinding head through a magnetic mounting seat, the miniature infrared temperature measurement array can be embedded in the edge of the bottom of the grinding head in order not to block the grinding area, and the vibration acceleration sensor can be pasted at the center position of the top of the grinding head.

[0020] It should be noted that the measurement accuracy of the triaxial force sensor is less than or equal to 0.1N, the sampling rate of the acoustic emission sensor is greater than or equal to 1MHz, the temperature measurement range of the miniature infrared temperature measurement array should meet 0℃-300℃, and the temperature measurement accuracy is not more than ±0.5℃, and the frequency response range of the vibration acceleration sensor is 10Hz-10kHz, thereby meeting the requirements of nanometer grinding industrial grade.

[0021] Further, the normal force, tangential force and moment data during the grinding process are obtained based on the triaxial force sensor as force signal data.

[0022] Specifically, the triaxial force sensor configured can collect force components along X, Y and Z three orthogonal directions and torque components around the three axes in real time during the grinding process, wherein the Z-direction force is the normal force, the X and Y-direction forces are the tangential forces, and the torque around the Z-axis is the grinding torque.

[0023] In some possible embodiments, assuming that the triaxial force sensor collects the Z-direction normal force, X-direction tangential force, Y-direction tangential force and torque around the Z-axis in real time during the grinding process of the sapphire substrate, outputs a group of data every 1ms, for example, collects the normal force of 12.5N, the X-direction tangential force of 3.2N, the Y-direction tangential force of 2.8N and the torque of 0.5Nm at a certain moment. The above data is the force signal data at the current moment.

[0024] Further, the acoustic emission raw signal during the grinding process is obtained based on the acoustic emission sensor as the acoustic emission signal data.

[0025] Specifically, the acoustic emission sensor configured obtains the elastic wave signals generated by the micro events such as abrasive cutting into the workpiece, abrasive wear and interface friction during the grinding process, and the raw signal is a continuous time domain voltage signal.

[0026] In some possible embodiments, during the grinding of the sapphire substrate, the abrasive cutting into the sapphire crystal generates high-frequency transient acoustic emission signals, the friction vibration generates medium-frequency continuous signals, and the structure resonance of the grinding head generates low-frequency signals. The acoustic emission sensor converts these signals into voltage signals, for example, 1000 voltage data points are collected in a certain 1ms time period to form a continuous time domain raw signal, and the continuous time domain raw signal is the acoustic emission signal data.

[0027] Further, the temperature distribution data during the grinding process is obtained based on the miniature infrared temperature measurement array as the temperature signal data.

[0028] Specifically, the miniature infrared temperature measurement array configured is composed of a plurality of infrared temperature measurement pixels, which can measure the temperature of different areas of the grinding interface at the same time to form a two-dimensional temperature distribution matrix.

[0029] It should be noted that the field of view angle of the micro infrared temperature measurement array needs to cover the entire grinding area, the temperature measurement frequency can be set to 10Hz, the temperature data consistent with the number of pixel arrays is output, and the temperature distribution graph of the grinding interface is obtained through interpolation processing, so as to reflect the spatial distribution characteristics of heat dissipation.

[0030] In some possible embodiments, during the grinding of the sapphire substrate, the temperature of the grinding interface is obtained using a micro infrared temperature measurement array such as a 32x32 pixel array, in the temperature distribution matrix collected at a certain moment, the temperature of the central grinding area is 85℃, and the temperature of the edge area is 62℃, through interpolation processing, a temperature distribution graph with the center of the substrate as a high temperature area and a gradient decrease towards the edge is obtained, and the temperature distribution matrix and the interpolated temperature graph are temperature signal data.

[0031] Further, the vibration amplitude and the vibration frequency of the grinding head during the grinding process are obtained based on the vibration acceleration sensor as vibration signal data.

[0032] Specifically, the vibration acceleration of the grinding head during the grinding process can be measured by the configured vibration acceleration sensor, and the vibration acceleration can be converted into vibration speed and displacement, i.e., amplitude, through integral operation, and the vibration frequency component can be extracted through Fourier transform.

[0033] It should be noted that the collection direction of the vibration acceleration sensor needs to be consistent with the direction of the force sensor, the collection frequency can be set to 1kHz, the output data is three-axis acceleration value, and the vibration amplitude and the peak frequency are obtained through signal processing.

[0034] In some possible embodiments, during the grinding of the sapphire substrate, the vibration acceleration sensor collects a Z-axis acceleration peak value of 12m / s², the vibration amplitude is 3.5μm obtained through integral calculation, and after Fourier transform is performed on the acceleration signal, the peak frequency is 250Hz, and the vibration amplitudes of the X-axis and the Y-axis are 1.8μm and 2.1μm respectively, and the peak frequencies are 220Hz and 235Hz respectively, and the above vibration amplitude and frequency data are vibration signal data.

[0035] Further, the force signal data, the acoustic emission signal data, the temperature signal data, and the vibration signal data are synchronized, aligned, and denoised for preprocessing to obtain grinding data.

[0036] Specifically, the collection time of a certain sensor can be used as a reference for synchronization and alignment, the time axes of the four types of signal data are unified through timestamp matching, and it is ensured that the grinding data at the same moment corresponds.

[0037] It can be understood that in the nanometer grinding process, the force signal data, the acoustic emission signal data, the temperature signal data and the vibration signal data are key indicators reflecting the grinding process state; the grinding force directly affects the material removal rate and the surface quality, and excessive or uneven force will cause workpiece deformation or surface damage; the acoustic emission signal can capture the microscopic phenomena such as material fracture and abrasive wear in the grinding process, and can detect abnormal wear or crack initiation in advance; the temperature change is related to the grinding heat, and excessive temperature will cause workpiece thermal deformation, affect the machining precision, and may also cause the change of material performance; the vibration signal can reflect the dynamic stability of the grinding system, and if there is abnormal vibration, there may be equipment failure or unreasonable grinding parameters; therefore, in this embodiment, the four types of data are acquired to comprehensively monitor the grinding process from multiple dimensions and provide a reliable basis for adaptive adjustment of process parameters.

[0038] It should be noted that different signals can be denoised and preprocessed using differentiated methods, such as sliding average filtering for force signals, wavelet threshold denoising for acoustic emission signals, median filtering for temperature signals, and Kalman filtering for vibration signals to remove environmental interference and sensor noise and obtain uniform format grinding data.

[0039] In some possible embodiments, taking a certain 10s processing period of sapphire substrate grinding as an example, the time axes of the force signal, the acoustic emission signal, the temperature signal and the vibration signal are unified through time stamping based on the force signal data collection time, to ensure that the force data, the acoustic emission data, the vibration data and the temperature data at the same time correspond; then the force signal can be subjected to sliding average filtering to remove transient pulse noise, the acoustic emission signal can be subjected to wavelet denoising to remove environmental electromagnetic interference, the temperature signal can be subjected to median filtering to remove isolated hot spot noise, and the vibration signal can be subjected to Kalman filtering to reduce structural inherent noise, to finally obtain the grinding data.

[0040] In step S2, feature extraction is performed on the grinding data to obtain feature parameters, and the grinding state is obtained in combination with empirical wavelet transform.

[0041] Further, the acoustic emission signal data in the grinding data is subjected to multi-scale decomposition based on the empirical wavelet transform to obtain abrasive cutting component, friction vibration component and structure resonance component.

[0042] Specifically, the Fourier spectrum of the acoustic emission signal in the grinding data is first segmented to obtain the frequency boundary of each component, and then the acoustic emission signal is decomposed into components of different frequency scales through empirical wavelet transform, wherein the abrasive cutting component corresponds to a high frequency band and is a transient pulse signal; the friction vibration component corresponds to a medium frequency band and is a continuous oscillation signal; and the structure resonance component corresponds to a low frequency band and is a low amplitude stationary signal.

[0043] In some possible embodiments, before the empirical wavelet transform of the acoustic emission signal data of the sapphire substrate grinding is performed, Fourier spectrum analysis of the original signal is performed, assuming that the high frequency band boundary is 500 kHz and the medium frequency band boundary is 100 kHz, three wavelet filters are constructed; then the original signal is decomposed into three components through the filter bank, the high frequency component of 500 kHz-1 MHz corresponds to the abrasive cutting component, which is represented as multiple sharp pulse; the medium frequency component of 100 kHz-500 kHz corresponds to the friction vibration component, which is represented as continuous sinusoidal oscillation; the low frequency component of 10 kHz-100 kHz corresponds to the structure resonance component, which is represented as low amplitude smooth fluctuation.

[0044] Further, the abrasive cutting component, the friction vibration component and the structure resonance component are feature extracted to obtain the energy proportion of the abrasive cutting component, the main frequency offset of the friction vibration component and the peak amplitude of the structure resonance component.

[0045] Specifically, the energy proportion of the abrasive cutting component is obtained by calculating the ratio of the total energy of the component to the total energy of the acoustic emission signal, the number of effective cutting abrasive is reflected by the energy proportion of the abrasive cutting component, the higher the energy proportion is, the more effective abrasive is; the main frequency offset of the friction vibration component is obtained by comparing the difference between the main frequency of the current component and the main frequency of the initial state, the degree of abrasive dulling is reflected by the main frequency offset of the friction vibration component, the more the main frequency shifts to low frequency, the duller the abrasive is; the peak amplitude of the structure resonance component is obtained by extracting the maximum amplitude of the structure resonance component, the contact stiffness of the grinding interface is reflected by the peak amplitude, the greater the amplitude is, the higher the contact stiffness is.

[0046] In some possible embodiments, the empirical wavelet transform of the sapphire substrate grinding is performed, the total energy of the abrasive cutting component is 120 μJ, the total energy of the acoustic emission signal is 200 μJ, the energy proportion is calculated to be 60%; the main frequency of the friction vibration component in the initial state is 350 kHz, the current main frequency is 320 kHz, the main frequency offset is obtained to be-30 kHz; the maximum amplitude of the structure resonance component is 0.8 V, that is, the peak amplitude is 0.8 V.

[0047] Further, the force signal data in the grinding data is feature extracted to obtain the peak force, the average force and the force fluctuation coefficient.

[0048] Specifically, the maximum value of the force signal in a certain statistical window is represented as the peak force, the instantaneous maximum cutting load is reflected by the peak force; the arithmetic mean of the force signal in the statistical window is the average force, the stable cutting load is reflected by the average force; the ratio of the standard deviation of the force signal in the statistical window to the average force is the force fluctuation coefficient, the load stability is reflected by the force fluctuation coefficient, the smaller the fluctuation coefficient is, the more stable the load is.

[0049] In some possible embodiments, the sapphire substrate grinding 1s is taken as a statistical window, a feature is extracted from the normal force signal data, the maximum value of the normal force in the window is 15N, the arithmetic mean of all data is 12N, the standard deviation is 0.8N, and the force fluctuation coefficient is calculated as 0.8 / 12≈0.067; similarly, the peak force of the X-direction tangential force is 4.5N, the average force is 3.2N, and the force fluctuation coefficient is 0.05, the peak force of the Y-direction tangential force is 4.2N, the average force is 2.8N, and the force fluctuation coefficient is 0.045, so as to obtain the feature parameter set of the force signal.

[0050] Further, the temperature signal data in the grinding data is characterized to obtain the average temperature and the temperature gradient.

[0051] Specifically, the arithmetic mean of the temperature distribution in a certain statistical window is taken as the average temperature, and the average temperature is used to reflect the overall heat dissipation level; the ratio of the maximum temperature difference in the statistical window to the radius of the grinding area is the temperature gradient, and the temperature gradient is used to reflect the unevenness of the temperature distribution, and the greater the gradient, the more uneven the temperature distribution.

[0052] It should be noted that before the feature extraction, the temperature distribution data needs to be subjected to outlier rejection, such as eliminating isolated points exceeding 3 times the standard deviation, so as to avoid the influence of abnormal values on the average temperature and the temperature gradient.

[0053] In some possible embodiments, in the 1s statistical window of the sapphire substrate grinding, 10 groups of temperature distribution data are obtained, and after eliminating 1 group of abnormally high temperature data, such as temperature 120℃ exceeding 3 times the standard deviation, the average temperature of the remaining 9 groups of data is 78℃; the maximum temperature difference of each group of temperature distribution is 25℃ on average, assuming that the radius of the grinding area is 5cm, the temperature gradient is calculated as 25℃ / 5cm=5℃ / cm, and the average temperature 78℃ and the temperature gradient 5℃ / cm are the feature parameters of the temperature signal.

[0054] Further, the vibration signal data in the grinding data is characterized to obtain the root mean square value and the peak frequency.

[0055] Specifically, the root mean square of the vibration acceleration signal in a certain statistical window is used to obtain the vibration root mean square value, and the vibration root mean square value is used to reflect the overall intensity of the vibration, and the greater the vibration root mean square value, the more violent the vibration; the peak frequency is obtained by taking the frequency corresponding to the peak of the Fourier spectrum of the vibration acceleration signal in the statistical window. In some possible embodiments, the vibration signal data in the sapphire substrate grinding window is extracted to obtain the feature, the root mean square value of the Z-axis vibration acceleration is 8.5 m / s2, the X-axis is 5.2 m / s2, and the Y-axis is 6.1 m / s2, and the maximum value 8.5 m / s2 is taken as the vibration root mean square value; the Fourier transform is performed on the Z-axis vibration signal, and the frequency corresponding to the frequency spectrum peak value is 250 Hz, so that the peak frequency is 250 Hz, and the feature parameter of the vibration signal is formed.

[0056] Further, a mapping relationship between the feature parameter and the grinding state is established, and the current grinding state is obtained based on the mapping relationship.

[0057] Specifically, the mapping relationship between the feature parameter and the grinding state is constructed through historical experimental data, and a mathematical correlation between the feature parameter and the grinding microstate is established by using multivariate regression analysis. For example, when the energy proportion of the abrasive grain cutting component is greater than or equal to 50%, the force fluctuation coefficient is less than or equal to 0.1, and the vibration root mean square value is less than or equal to 10 m / s2, the corresponding grinding state is sufficient effective cutting; when the main frequency offset of the friction component is less than or equal to -20 kHz, and the temperature gradient is greater than or equal to 4 ℃ / cm, the corresponding grinding state is light dulling of the abrasive grain; when the peak amplitude of the structure resonance component is greater than or equal to 1.0 V, and the average force of the normal force is greater than or equal to 15 N, the corresponding grinding state is that the interface contact is too tight.

[0058] In some possible embodiments, the real-time feature parameters of the sapphire substrate grinding, such as the energy proportion of the abrasive grain cutting component 60%, the main frequency offset of the friction component -30 kHz, the peak amplitude of the structure resonance component 0.8 V, the average force of the normal force 12 N, the force fluctuation coefficient 0.067, the temperature gradient 5 ℃ / cm, and the vibration root mean square value 8.5 m / s2, are substituted into the mapping relationship, and the determination result is that the effective cutting is sufficient and the abrasive grain is slightly dull, that is, the current grinding state is the effective cutting state of the slightly dull abrasive grain.

[0059] In step S3, the grinding state is identified based on the topological data analysis and the hidden Markov model, and an identification result is obtained.

[0060] Further, a high-dimensional feature vector is constructed based on the feature parameter, and the topological data analysis is used to process the high-dimensional feature vector to obtain a process state category.

[0061] Specifically, after all the feature parameters extracted in step S2 are normalized, a multi-dimensional high-dimensional feature vector is constructed; and by using the topological data analysis method of the persistent homology method, the high-dimensional feature vector is mapped to a point cloud, the persistent homology of the Rips complex is calculated, and the clustering cluster corresponding to the long-life feature in the point cloud is identified, and each clustering cluster corresponds to a stable process state category.

[0062] It should be noted that the feature with a duration greater than or equal to 5 s is defined as a long-life feature.

[0063] It can be understood that the topological data analysis is a geometric data analysis method capable of capturing the global structure and internal topological characteristics of data; in the embodiment, similar process states will be shown as clustering clusters with long-life characteristics on the topological structure after the characteristic parameters are mapped to the topological space through the topological data analysis; compared with the traditional threshold or empirical rule data analysis method, the method using the topological data analysis in the embodiment can better obtain the potential law of the process state, and provides an accurate state classification basis for process self-adaptive adjustment.

[0064] In some possible embodiments, 1000 groups of 12-dimensional feature vectors continuously collected during the sapphire substrate grinding process are normalized to construct a high-dimensional point cloud, wherein the dimensions of the feature vectors are determined according to the grinding data of step S2; three clustering clusters corresponding to three long-life characteristics are identified through the persistent homology calculation of the topological data analysis, including a high-efficiency cutting state corresponding to cluster 1, a stable grinding state corresponding to cluster 2, and a critical wear state corresponding to cluster 3, that is, three process state categories.

[0065] Further, the process state categories are used as a hidden state set of the hidden Markov model, and the feature parameter sequence is used as an observation sequence, and the hidden Markov model is trained to obtain a state transition probability matrix and an observation probability matrix.

[0066] Specifically, the process state categories are used as a hidden state set of the hidden Markov model, and the feature parameter sequence is used as an observation sequence, and the hidden Markov model is trained in combination with the Baum-Welch algorithm, a plurality of complete feature parameter sequences collected in the historical grinding process are input, and a state transition probability matrix and an observation probability matrix are obtained through iterative optimization.

[0067] It should be noted that the hidden Markov model is a model used for statistics, commonly used for processing dynamic systems with hidden states and observable sequences, which assumes that the system has a set of hidden states, which cannot be directly observed but can be indirectly inferred through the observation sequence; in the embodiment, the hidden Markov model describes the conversion law between the hidden states through the state transition probability matrix, and establishes the probability relationship between the hidden states and the observation sequence through the observation probability matrix.

[0068] It can be understood that the Baum-Welch algorithm is a training method of the hidden Markov model, which is an iterative algorithm based on the expectation maximization principle; in the case of known observation sequence but unknown hidden state, the algorithm continuously iteratively optimizes the state transition probability matrix and the observation probability matrix in the embodiment, so that the probability of the model generating a given observation sequence is maximized.

[0069] In some possible embodiments, the historical data of sapphire substrate grinding is taken as a training set, 100 groups of feature parameter sequences are input, the probability of efficient cutting state to stable grinding state in the state transition probability matrix is 0.7, the probability of stable grinding state to critical wear state is 0.3, and the probability of critical wear state to stable grinding state is 0.1; in the observation probability matrix, the observation probability of the abrasive particle cutting component energy proportion being greater than or equal to 50% in the efficient cutting state is 0.95, the observation probability of the friction component main frequency offset being-20 to-40 kHz in the stable grinding state is 0.85, and the observation probability of the vibration root mean square value being greater than or equal to 15 m / s² in the critical wear state is 0.9.

[0070] Further, the real-time extracted feature parameters are input into the hidden Markov model, the posterior probability of the current state and the current grinding state category are obtained based on the forward algorithm, the confidence of the grinding state category is obtained based on the posterior probability, and the grinding state category and the corresponding confidence are taken as the recognition result.

[0071] The process state category includes an efficient cutting state, a stable grinding state and a critical wear state.

[0072] Specifically, the real-time feature parameter sequence is input into the trained hidden Markov model, and the posterior probability of each hidden state at each time step is calculated by using the forward algorithm, that is, the probability of being in each state is given the current and historical observation sequence; the hidden state with the maximum posterior probability is selected as the current grinding state category; the confidence is the normalized value of the maximum posterior probability, and the value range is 0-1, and the greater the value is, the more reliable the state recognition is.

[0073] It should be noted that, in order to avoid state misjudgment caused by instantaneous noise, when the posterior probability of a certain state is the maximum for 3 consecutive time steps, the state is determined as the current state category, so as to ensure the stability of the recognition result.

[0074] In some possible embodiments, during the real-time grinding of the sapphire substrate, a group of 12-dimensional feature parameters is extracted every 10 ms, and the hidden Markov model is continuously input, it is found that at the t th time, the posterior probability of the efficient cutting state is 0.2, the posterior probability of the stable grinding state is 0.75, and the posterior probability of the critical wear state is 0.05, the state corresponding to the maximum posterior probability is the stable grinding state; at the t+1 th and t+2 th time, the posterior probabilities of the stable grinding state are 0.78 and 0.82 respectively, and the maximum is maintained for 3 consecutive time steps, so it is determined that the current grinding state category is the stable grinding state, and the confidence is 0.82, and finally the recognition result is the stable grinding state, and the confidence is 0.82.

[0075] In step S4, energy flow data is obtained, a process entropy production rate is obtained based on the energy flow data, performance evaluation is performed based on the recognition result and the process entropy production rate, and an evaluation result is obtained.

[0076] Further, the spindle motor power consumption and the pressure system power consumption are acquired, and a material removal energy is acquired based on an abrasive grain cutting component energy of the acoustic emission signal data, and the spindle motor power consumption, the pressure system power consumption and the material removal energy are composed into energy flow data.

[0077] Specifically, the spindle motor power consumption and the pressure system power consumption are acquired in real time, the spindle motor power consumption directly reflects energy input for driving the grinding head to rotate, and the pressure system power consumption reflects energy input for maintaining the grinding pressure; the material removal energy is acquired by multiplying the abrasive grain cutting component energy and a calibration coefficient, the calibration coefficient is acquired through historical data, and the calibration coefficients of different grinding materials are different; and the spindle motor power consumption, the pressure system power consumption and the material removal energy are composed into the energy flow data.

[0078] In some possible embodiments, it is assumed that, in a 10s statistical window of sapphire substrate grinding, the average spindle motor power consumption is 150W, the average pressure system power consumption is 30W, the total energy of the abrasive grain cutting component of the acoustic emission signal is 1410J, the calibration coefficient of the sapphire is 0.85, the material removal energy is 0.85*1410J≈1200J, and the energy flow data is [spindle power consumption 150W, pressure power consumption 30W, material removal energy 1200J].

[0079] Further, the spindle motor power consumption and the pressure system power consumption are summed to acquire input power data, and the material removal energy is used as useful power data.

[0080] Specifically, the input power data is the sum of the spindle motor power consumption and the pressure system power consumption in the statistical window, the total energy input of the grinding process is reflected through the input power data, and the average power consumption needs to be multiplied by the statistical window length when calculating; the useful power data is the material removal energy, and the energy used for actual material cutting and removal is reflected through the useful power.

[0081] It should be noted that the difference between the input power and the useful power is the theoretical total dissipated power, which is used to verify the directly measured dissipated power, if the verification deviation is less than 10%, the verification is qualified, if the verification deviation is greater than or equal to 10%, otherwise, data calibration is triggered.

[0082] In some possible embodiments, the 10s statistical window of the sapphire substrate grinding is continued, the average power consumption of the spindle is 150W, the average power consumption of the pressure system is 30W, the input power is (150+30)Wx10s=1800J; assuming that the actual material removal energy is 1200J, the theoretical total dissipated power is 1800J-1200J=600J; the total dissipated power directly measured is obtained for verification, assuming that the total dissipated power directly measured is 580J, the deviation is within 10%, and the data is valid; if the total dissipated power directly measured is 500J, the deviation is more than 10%, the device idle loss needs to be separated according to the calibration mechanism, and the total dissipated power in the theoretical case is recalculated, that is, the process part in the input power is subtracted from the useful work.

[0083] Further, the processing time of the nanometer grinding process is obtained, and the process entropy production rate is obtained based on the input power data, the useful work data and the processing time.

[0084] Specifically, the processing time of the nanometer grinding process is obtained, the input power data, the useful work data and the processing time are input into the process entropy production rate formula to obtain the process entropy production rate, and the process entropy production rate formula is specifically represented as , the process entropy production rate is represented as the process entropy production rate, the input power data is represented as the input power data, the useful work data is represented as the useful work data, and the processing time of the nanometer grinding process is represented as the processing time of the nanometer grinding process. , the process entropy production rate is represented as the process entropy production rate, the input power data is represented as the input power data, the useful work data is represented as the useful work data, and the processing time of the nanometer grinding process is represented as the processing time of the nanometer grinding process. , the process entropy production rate is represented as the process entropy production rate, the input power data is represented as the input power data, the useful work data is represented as the useful work data, and the processing time of the nanometer grinding process is represented as the processing time of the nanometer grinding process. , the process entropy production rate is represented as the process entropy production rate, the input power data is represented as the input power data, the useful work data is represented as the useful work data, and the processing time of the nanometer grinding process is represented as the processing time of the nanometer grinding process. , the process entropy production rate is represented as the process entropy production rate, the input power data is represented as the input power data, the useful work data is represented as the useful work data, and the processing time of the nanometer grinding process is represented as the processing time of the nanometer grinding process.

[0085] In some possible embodiments, the 10s statistical window of the sapphire substrate grinding is continued, the input power data is 1800J, the useful work data is 1200J, and the process entropy production rate is calculated to be (1800-1200) / 10=60J / s, so that the process entropy production rate of the 10s statistical window is 60J / s.

[0086] Further, a performance evaluation index system is set, and performance evaluation is performed in combination with the identification result to obtain an evaluation result.

[0087] Further, the material removal amount is obtained, the material removal rate is obtained based on the material removal amount and the processing time, the material removal rate is taken as a processing efficiency index, the surface shape error of the processed material is obtained, the surface shape error is taken as a processing precision index, the process entropy production rate is taken as a process stability index, and the performance evaluation index system is set based on the processing efficiency index, the processing precision index and the process stability index.

[0088] Specifically, the mass difference of the material before and after processing is obtained, and the volume removal amount of the material is calculated by combining the density of the material, and then the material removal rate is obtained according to the material removal volume / processing time, and the material removal rate is used to reflect the material removal capacity per unit time; the surface shape error of the material, that is, the PV value, can be measured in real time by a white light interferometer, and the surface shape error is used to reflect the flatness deviation of the workpiece surface, and finally the process entropy production rate is used as a process stability index, the lower the process entropy production rate, the lower the process disorder degree, and the better the stability, and the performance evaluation index system is set by the processing efficiency index, the processing precision index and the process stability index.

[0089] It should be noted that the performance evaluation index system adopts a weighted scoring method, and the weights in the weighted scoring method are dynamically adjusted according to the current grinding state category, including, for example, in the high-efficiency cutting state, the processing efficiency index weight is 0.4, the processing precision index weight is 0.3, and the process stability index weight is 0.3; in the stable grinding state, the processing efficiency index weight is 0.3, the processing precision index weight is 0.4, and the process stability index weight is 0.3; in the critical wear state, the processing efficiency index weight is 0.2, the processing precision index weight is 0.3, and the process stability index weight is 0.5, so as to ensure that the evaluation meets the current process core demand.

[0090] In some possible embodiments, within a 1s window of sapphire substrate grinding, assuming that the material removal volume is 0.0628mm³, the material removal rate is 0.0628mm³ / s; the surface shape error obtained by the white light interferometer is 0.15μm; the process entropy production rate is 60J / s; the current recognition result is the stable grinding state, and therefore the index weight is set to efficiency 0.3, precision 0.4, and stability 0.3; each index can be standardized by, for example, a min-max normalization method, to obtain a standardized score, such as efficiency 62.8 points, precision 70 points, and stability 87.5 points.

[0091] Further, based on the performance evaluation index system, the grinding state category in the recognition result is scored to obtain a score result.

[0092] For example, the scores of each index of the grinding state category and the corresponding weights can be fused by a weighted summation method to obtain a final score result, and the weight setting of the stable grinding state is continued, and the weights of each index are 0.3, 0.4 and 0.3 respectively, and the score result is (62.8x0.3+ (70x0.4) + (87.5x0.3) = 73.09 points, that is, the score result of the current process performance is 73.09 points.

[0093] Further, a score threshold is set, and the score result is compared with the score threshold, and if the score result is greater than or equal to the score threshold, it is determined that the performance meets the standard.

[0094] Specifically, the score threshold is set according to the process requirements and combined with the grinding state category differentiation. The high-efficiency cutting state needs to consider both efficiency and accuracy, so the threshold is set to 80 points. The stable grinding state focuses on accuracy and stability, so the threshold is set to 70 points. The critical wear state prioritizes stability, and the threshold is set to 60 points. When the score result is greater than or equal to the threshold of the corresponding state, it means that the current process performance meets the expectations and no adjustment is needed.

[0095] In some possible embodiments, the current recognition result is the stable grinding state, and the corresponding score threshold is set to 70 points. The current score result is 73.09 points, which is greater than or equal to the score threshold of 70 points, so it is determined that the performance meets the standard, and the current process state is maintained to continue grinding.

[0096] Furthermore, if the score result is less than the score threshold, it is determined that the performance does not meet the standard.

[0097] When the score result is less than the score threshold of the corresponding state, the system marks that the performance does not meet the standard and records the short-board indicator, that is, the indicator with the lowest standardized score, to provide a direction for subsequent parameter optimization. For example, if the abrasive grains are severely dull due to the grinding, the material removal rate in the statistical window is 0.03 mm³ / s, the surface shape error is 0.2 μm, and the process entropy production rate is 120 J / s. The current recognition result is still the stable grinding state, and the score is reduced to 50 points, which is less than the score threshold of 70 points. It is determined that the performance does not meet the standard, and the short-board indicator is the machining efficiency. The system records all data of this substandard state and triggers the parameter optimization process.

[0098] In some possible embodiments, it is assumed that the sapphire substrate grinding is performed for 10 minutes, the abrasive grains are severely dull, the material removal rate in the statistical window is 0.03 mm³ / s, the surface shape error is 0.2 μm, and the process entropy production rate is 120 J / s. The current recognition result is still the stable grinding state, and the score is reduced to 50 points, which is less than the score threshold of 70 points. It is determined that the performance does not meet the standard, and the short-board indicator is the machining efficiency. The system records all data of this substandard state and triggers the parameter optimization process.

[0099] In step S5, the process parameters are optimized based on the alternating direction multiplier method to obtain optimized process parameters, and the process parameters are adaptively adjusted based on the optimized process parameters.

[0100] Further, historical production data is obtained, and the performance substandard state in the evaluation result is analyzed in combination with the historical production data to confirm a set of process parameters to be optimized, the set of process parameters to be optimized including a trajectory parameter and a machining parameter.

[0101] Specifically, the historical production data includes a corresponding relationship between the trajectory parameter, the machining parameter, and the machining efficiency, the accuracy, and the stability. The trajectory parameter includes a fractal dimension and an initial line segment length. The machining parameter includes a grinding pressure, a spindle speed, and a feed speed. Correlation analysis is performed by calculating a Pearson correlation coefficient to quantify the correlation strength between the parameter and the short-board indicator. If the absolute value of the correlation coefficient is greater than or equal to 0.6, it indicates that the parameter has a significant impact on the indicator, and the parameter belongs to the process parameters to be optimized.

[0102] It can be understood that in statistics, it is generally considered that the correlation coefficient with an absolute value between 0.5 and 0.7 is moderately correlated, and 0.6 is used as a threshold to effectively screen out key parameters that have a significant impact on the index, avoid the interference of too many weakly related parameters on the optimization direction, and prevent the threshold from being too high to miss potential important parameters.

[0103] It should be noted that if the machining efficiency is a short-board index, parameters with strong correlation with the material removal rate need to be screened out to ensure the accuracy of the optimization direction; the trajectory parameters indirectly affect the machining efficiency by affecting the coverage density and idle stroke proportion of the grinding trajectory; the machining parameters directly determine the cutting strength and speed of the abrasive particles.

[0109] In some possible embodiments, for the case that the grinding machining efficiency of the sapphire substrate does not meet the standard, 100 sets of historical data are retrieved for correlation analysis, the correlation coefficient of the grinding pressure and the material removal rate is 0.78, the spindle speed is 0.72, the fractal dimension is 0.65, the feed speed is 0.53, and the initial line segment length is 0.55; it is found that when the grinding pressure, spindle speed and fractal dimension are greater than or equal to this, the set of process parameters to be optimized is determined as the grinding pressure in the machining parameters and the fractal dimension in the trajectory parameters, to ensure that the optimization focuses on the key influencing factors.

[0105] Further, a double-layer nested optimization target is constructed, including an inner optimization target and an outer optimization target, the inner optimization target is to maximize the machining precision and efficiency based on the current trajectory parameters, and the outer optimization target is to minimize the process entropy production rate based on the process state.

[0106] Specifically, in order to take into account the local performance optimization and the global process stability, a double-layer nested optimization target is constructed, wherein the inner optimization focuses on the improvement of the machining precision and efficiency under the current trajectory parameters, and is realized by adjusting the grinding pressure, spindle speed and feed speed of the machining parameters, the objective function is to maximize the weighted sum of the machining precision score and the machining efficiency score, and the weights are allocated according to the importance of the machining progress and the machining efficiency. In this embodiment, the importance of the machining progress and the machining efficiency is the same, so the weights of the two are both set to 0.5, to ensure the improvement of the efficiency without sacrificing the precision; the outer optimization takes the minimization of the process entropy production rate as the core, optimizes the trajectory coverage density and idle stroke proportion by adjusting the fractal dimension in the trajectory parameters, reduces the energy dissipation, and the objective function is to minimize the process entropy production rate, to ensure the long-term stability of the process. The two layers of targets are solved by alternating iteration, to realize the collaborative optimization of the precision, efficiency and stability.

[0107] In some possible embodiments, for the case where the processing efficiency of the sapphire substrate does not meet the requirements and the current state is a steady grinding state, a double-layer optimization target is constructed, where the inner-layer optimization target is to maximize the weighted sum of the processing precision score and the efficiency score by adjusting the grinding pressure, the spindle speed, and the feed speed; and the outer-layer optimization target is to reduce the process entropy production rate from the current 90 J / s to below 80 J / s by adjusting the fractal dimension to optimize the trajectory coverage density and reduce the invalid energy consumption.

[0108] Further, the double-layer nested optimization target is decomposed into multiple sub-problems based on the alternating direction multiplier method, and the optimal solutions of the sub-problems are solved by alternating iteration.

[0109] Specifically, the alternating direction multiplier method decomposes the complex double-layer optimization problem into sub-problems that can be solved independently, thereby reducing the computational complexity. The decomposition logic is specifically that the inner-layer optimization target and the outer-layer optimization target are first split into two sub-problems, a penalty parameter and a dual variable are introduced to construct an augmented Lagrangian function, and the optimal solution is obtained by the process of alternating solving sub-problems, updating the dual variable, and iterative convergence. When solving the sub-problems, the inner-layer sub-problem fixes the trajectory parameters and optimizes the processing parameters; and the outer-layer sub-problem fixes the processing parameters and optimizes the trajectory parameters, to ensure the synergy of the two layers of targets.

[0110] It should be noted that the alternating direction multiplier method has a fast convergence speed, and is particularly suitable for real-time control in industrial scenarios. Further, by separating the inner-layer and outer-layer optimization tasks, the local optimal trap in the traditional global optimization can be effectively avoided, thereby improving the global optimality of the solution. The alternating direction multiplier method supports a distributed computing architecture. In the multi-device collaborative grinding scenario, each terminal can process the sub-problems in parallel, thereby improving the overall optimization efficiency and achieving intelligent level for the nanometer grinding process.

[0111] In some possible embodiments, for the double-layer optimization target of the sapphire substrate, the alternating direction multiplier method is used to decompose the target into two sub-problems, including defining the inner layer as sub-problem 1, fixing the fractal dimension as 1.6, optimizing the grinding pressure, the spindle speed, and the feed speed, and maximizing the weighted sum of the precision and the efficiency; defining the outer layer as sub-problem 2, fixing the current optimal solution of the processing parameters, optimizing the fractal dimension, and minimizing the process entropy production rate; setting the penalty parameter as 10 and the initial value of the dual variable as 0, and alternatingly solving sub-problem 1 and sub-problem 2, updating the dual variable every 10 iterations, until the convergence condition is met.

[0112] Further, in the solving process, the inner-layer optimization sub-problem satisfies the constraint condition that the trajectory coverage density is greater than or equal to a preset density threshold and the idle stroke proportion is less than a preset proportion threshold.

[0113] Specifically, the trajectory coverage density threshold ensures that all areas of the workpiece surface are covered by the grinding trajectory, avoiding precision defects caused by missed processing; the idle stroke proportion threshold limits the invalid motion time, ensuring that the processing efficiency is not excessively sacrificed.

[0114] It should be noted that the trajectory coverage density is calculated by the ratio of the trajectory length to the grinding area, and the idle stroke proportion is calculated by the ratio of the idle stroke time to the total processing time; the constraint condition adopts an inequality constraint form, which is embedded in the sub-problem solution of the alternating direction multiplier method, and if the optimization result violates the constraint, the parameter is adjusted by the Lagrange multiplier to force the constraint requirement to be met.

[0115] In some possible embodiments, for the inner layer sub-problem of sapphire substrate grinding, the preset trajectory coverage density threshold is greater than or equal to 95%, and the idle stroke proportion threshold is less than or equal to 5%; when optimizing the processing parameters, if the trajectory coverage density corresponding to the grinding pressure 18 N, the spindle speed 5500 r / min, and the feed speed 14 mm / min is 93%, and the idle stroke proportion is 6%, the group of parameters is determined as an invalid solution, and the feed speed is reduced to 12 mm / min by adjusting the Lagrange multiplier, so that the trajectory coverage density is increased to 96% and the idle stroke proportion is reduced to 4%, satisfying the constraint condition.

[0116] It can be understood that, because nanometer grinding has very high requirements on surface processing precision, if the coverage density is lower than the value, the workpiece surface is prone to have areas that are not fully ground, resulting in that key indicators such as surface roughness and flatness do not meet the standards, and affecting the functionality and service life of the product, therefore the trajectory coverage density threshold is set to 95%; considering factors such as equipment energy consumption, processing cost, and production cycle, a too high idle stroke proportion means that the equipment is doing a lot of movement, which not only increases the energy consumption cost, but also prolongs the overall processing time and reduces the production efficiency, therefore the idle stroke proportion threshold is set to 5%.

[0117] Further, the outer layer optimization sub-problem satisfies the constraint condition that the process entropy production rate is less than the entropy production rate threshold.

[0118] Specifically, the constraint condition of the outer layer optimization sub-problem focuses on process stability, and the entropy production rate threshold is calibrated according to historical optimal process data, ensuring that the optimized process is in a low-dissipation and high-stability state; the entropy production rate constraint adopts an inequality constraint, and if the entropy production rate corresponding to the fractal dimension after optimization is higher than the threshold, the parameter is adjusted by expanding or reducing the fractal dimension until the constraint is satisfied.

[0119] It should be noted that through historical data discovery, it is found that when the entropy production rate is below 80 J / s, the running vibration amplitude of the grinding equipment can be controlled within 0.05 mm, the process parameter fluctuation range is reduced to ±3%, and the particle size distribution variation coefficient of the grinding product is stable below 5%, which can reduce the risk of equipment wear and product quality fluctuation caused by excessive energy dissipation; Therefore, 80 J / s is set as the critical value of process stability. The threshold value calibrated in this way can ensure that the optimized process is in a low dissipation and high stability state.

[0120] In some possible embodiments, for the outer layer sub-problem of the sapphire substrate, the preset entropy production rate threshold is less than or equal to 80 J / s; when the optimized fractal dimension is 1.8, the calculated process entropy production rate is 85 J / s, the fractal dimension needs to be further adjusted, if it is increased to 1.9, the trajectory coverage density is increased, the energy dissipation is increased, and the entropy production rate is increased to 88 J / s; if it is reduced to 1.7, the trajectory is more compact, the energy dissipation is reduced, and the entropy production rate is reduced to 78 J / s, so the fractal dimension of 1.7 is determined as the effective solution.

[0121] Further, the change threshold is set, when the process parameter change amount of the adjacent two iterations is less than the change threshold, it is determined that the convergence is obtained, the optimal trajectory parameter and the optimal processing parameter are obtained, and the optimization process parameter is formed.

[0122] Specifically, the change threshold is set, the relative change amount of the process parameters of the adjacent two iterations is calculated, if the relative change amount of all parameters is less than the threshold, it is determined that the algorithm converges, and the current parameter is the optimal solution; At the same time, if the iteration number reaches the preset maximum number and still does not converge, such as 50 times, the parameter of the last iteration is taken as a suboptimal solution, so as to ensure that the system does not fall into infinite iteration.

[0123] It should be noted that in the nanoscale processing scene, the change threshold is too high, which may cause the process parameters to not reach the optimal state, affecting the grinding surface quality and processing precision; while a too low threshold will greatly increase the iteration number of the algorithm, prolong the calculation time and reduce the production efficiency; The threshold of 0.5% can ensure that the process parameters converge in a relatively stable and near-optimal solution range, and balance the consumption of calculation resources, so that the self-adaptive adjustment system has good practicability and response speed in actual production; Therefore, the change threshold is set to 0.5%.

[0124] In some possible embodiments, it is assumed that the parameters of the 30th iteration are the grinding pressure 16 N, the spindle speed 5200 r / min, the feed speed 13 mm / min, and the fractal dimension 1.7; the parameters of the 31st iteration are the grinding pressure 16.05 N, the spindle speed 5220 r / min, the feed speed 13.06 mm / min, and the fractal dimension 1.701; the relative change amounts of the grinding pressure, the spindle speed, the feed speed, and the fractal dimension are 0.31%, 0.38%, 0.46%, and 0.06% respectively, all of which are less than 0.5%, it is determined that the algorithm converges, and the optimal parameter combination is [grinding pressure 16.05 N, spindle speed 5220 r / min, feed speed 13.06 mm / min, and fractal dimension 1.7], and the composition is the optimized process parameter.

[0125] Further, a control instruction is generated based on the optimized process parameter, and the control instruction is sent to the grinding device, and the grinding device performs adaptive adjustment based on the control instruction.

[0126] For example, for the optimized process parameter of the sapphire substrate, the control instruction includes the trajectory interpolation instruction for the fractal dimension 1.7, and the initial line segment length 0.3 mm, and is sent to the multi-axis motion controller to drive the grinding head to generate a corresponding trajectory; and the processing driving instruction is the grinding pressure 16.05 N, the spindle speed 5220 r / min, and the feed speed 13.06 mm / min, and is sent to the pressure control system, the spindle motor driver, and the feeding system.

[0127] Figure 3 A schematic diagram of a self-adaptive adjustment system of a nano-grinding process is shown, which can implement the idea of the application.

[0128] Specifically, a self-adaptive adjustment system of a nano-grinding process includes: An acquisition module is configured to acquire force signal data, acoustic emission signal data, temperature signal data, and vibration signal data in a nano-grinding process, and compose grinding data, and acquire energy flow data; An extraction module is configured to perform feature extraction on the grinding data, acquire feature parameters, and acquire a grinding state by using an empirical wavelet transform; An identification module is configured to identify the grinding state based on topological data analysis and a hidden Markov model, and acquire an identification result; An evaluation module is configured to acquire a process entropy production rate based on the energy flow data, perform performance evaluation based on the identification result and the process entropy production rate, and acquire an evaluation result; An optimization module is configured to optimize process parameters based on an alternating direction multiplier method, and acquire optimized process parameters; An adjustment module is configured to perform adaptive adjustment based on the optimized process parameters.

[0129] The specific use and role of the embodiment are described below: First, the force signal data, acoustic emission signal data, temperature signal data and vibration signal data are acquired and combined into grinding data, then the grinding data is subjected to feature extraction to obtain feature parameters, and the grinding state is obtained in combination with the empirical wavelet transform, then the grinding state is identified through topological data analysis and a hidden Markov model to obtain an identification result, then the energy flow data is acquired, the process entropy production rate is obtained according to the energy flow data, the performance is evaluated through the identification result and the process entropy production rate to obtain an evaluation result, finally the process parameters are optimized through the alternating direction multiplier method to obtain optimized process parameters, and adaptive adjustment is performed according to the optimized process parameters, so that the intelligent requirements of adaptive regulation and control in the grinding process are realized through the technical solution based on accurate identification of the microscopic state and intelligent optimization of the process parameters.

[0130] It should be understood that the size of the serial number of each process described above does not mean the order of execution in the embodiments of the present application, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0131] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. An adaptive adjustment method for a nano-grinding process, characterized in that, It includes the following steps: In the nano-grinding process, force signal data, acoustic emission signal data, temperature signal data, and vibration signal data are acquired and combined to form grinding data; Feature extraction is performed on the grinding data to obtain feature parameters, and the grinding state is obtained by combining empirical wavelet transform. The grinding state is identified based on topological data analysis and hidden Markov models, and the identification results are obtained. Acquire energy flow data, obtain process entropy yield based on energy flow data, perform performance evaluation based on identification results and process entropy yield, and obtain evaluation results; The process parameters are optimized using the alternating direction multiplier method to obtain optimized process parameters, and adaptive adjustments are made based on these optimized process parameters.

2. The adaptive adjustment method for a nano-grinding process according to claim 1, characterized in that, In the nano-grinding process, force signal data, acoustic emission signal data, temperature signal data, and vibration signal data are acquired and combined to form grinding data, including: The grinding head is equipped with a triaxial force sensor, an acoustic emission sensor, a miniature infrared temperature measurement array, and a vibration acceleration sensor. The normal force, tangential force, and torque data during the grinding process are acquired using a triaxial force sensor as force signal data. Acoustic emission signals during the grinding process are obtained using acoustic emission sensors as acoustic emission signal data. Temperature distribution data during the grinding process is acquired using a miniature infrared thermometer array as temperature signal data. The vibration amplitude and frequency of the grinding head during the grinding process are obtained using a vibration acceleration sensor as vibration signal data. The force signal data, acoustic emission signal data, temperature signal data, and vibration signal data are synchronized, aligned, and denoised before preprocessing to obtain grinding data.

3. The adaptive adjustment method for a nano-grinding process according to claim 1, characterized in that, Feature extraction is performed on the grinding data to obtain feature parameters, and the grinding state is obtained by combining empirical wavelet transform, including: Based on empirical wavelet transform, the acoustic emission signal data in the grinding data is decomposed into multi-scale components to obtain the abrasive particle penetration component, friction vibration component and structural resonance component. Feature extraction is performed on the abrasive particle penetration component, friction vibration component, and structural resonance component to obtain the energy ratio of the abrasive particle penetration component, the main frequency offset of the friction vibration component, and the peak amplitude of the structural resonance component. Feature extraction is performed on the force signal data in the grinding data to obtain peak force, average force, and force fluctuation coefficient; Feature extraction is performed on the temperature signal data in the grinding data to obtain the average temperature and temperature gradient; Feature extraction is performed on the vibration signal data in the grinding data to obtain the root mean square value and peak frequency; Establish a mapping relationship between feature parameters and grinding state, and obtain the current grinding state based on the mapping relationship.

4. The adaptive adjustment method for a nano-grinding process according to claim 1, characterized in that, The grinding state is identified based on topological data analysis and hidden Markov models, and the identification results are obtained, including: A high-dimensional feature vector is constructed based on the feature parameters, and topological data analysis is used to process the high-dimensional feature vector to obtain the process state category; Using the process state category as the set of hidden states of the Hidden Markov Model and the feature parameter sequence as the observation sequence, the Hidden Markov Model is trained to obtain the state transition probability matrix and the observation probability matrix. The real-time extracted feature parameters are input into the Hidden Markov Model. The posterior probability of the current state and the current grinding state category are obtained based on the forward algorithm. The confidence level of the grinding state category is obtained based on the posterior probability. The grinding state category and the corresponding confidence level are used as the recognition result. The process state categories include high-efficiency cutting state, stable grinding state, and critical wear state.

5. The adaptive adjustment method for a nano-grinding process according to claim 1, characterized in that, Acquire energy flow data, obtain process entropy yield based on energy flow data, perform performance evaluation based on identification results and process entropy yield, and obtain evaluation results, including: The power consumption of the spindle motor and the power consumption of the pressure system are obtained. The material removal energy is obtained based on the abrasive cutting component energy of the acoustic emission signal data. The power consumption of the spindle motor, the power consumption of the pressure system, and the material removal energy are combined into energy flow data. The input power data is obtained by summing the power consumption of the spindle motor and the power consumption of the pressure system, and the material removal energy is used as the useful power data. The processing time of the nano-grinding process is obtained, and the entropy yield of the process is obtained based on the input work data, useful work data, and processing time. Establish a performance evaluation index system, and conduct performance evaluation based on the identification results to obtain the evaluation results.

6. The adaptive adjustment method for a nano-grinding process according to claim 5, characterized in that, Establish a performance evaluation index system, and conduct performance evaluation based on the identification results to obtain the evaluation results, including: The material removal amount is obtained, and the material removal rate is obtained based on the material removal amount and processing time. The material removal rate is used as a processing efficiency index. The surface shape error of the processed material is obtained and used as a processing accuracy index. The process entropy yield is used as a process stability index. A performance evaluation index system is set based on the processing efficiency index, processing accuracy index, and process stability index. Based on the performance evaluation index system, a score is obtained by combining the grinding state category in the identification results; Set a scoring threshold, compare the scoring results with the scoring threshold, and if the scoring result is greater than or equal to the scoring threshold, the performance is deemed to meet the standard; If the score is less than the score threshold, the performance is deemed substandard.

7. The adaptive adjustment method for a nano-grinding process according to claim 1, characterized in that, The process parameters are optimized using the alternating direction multiplier method to obtain optimized process parameters. Adaptive adjustments are then made based on these optimized process parameters, including: Historical production data is acquired, and based on the substandard performance status in the evaluation results, the historical production data is analyzed to identify the set of process parameters to be optimized. The set of process parameters to be optimized includes trajectory parameters and processing parameters. A two-layer nested optimization objective is constructed, including an inner optimization objective and an outer optimization objective. The inner optimization objective is to maximize processing accuracy and efficiency based on the current trajectory parameters, and the outer optimization objective is to minimize process entropy productivity based on the process state. The double-nested optimization objective is decomposed into multiple sub-problems based on the alternating direction multiplier method, and the optimal solution of each sub-problem is solved by alternating iteration. Set a threshold for the amount of change. When the change in process parameters between two consecutive iterations is less than the threshold, convergence is determined, and the optimal trajectory parameters and optimal processing parameters are obtained and combined into optimized process parameters. Control commands are generated based on optimized process parameters and sent to the grinding equipment, which then makes adaptive adjustments based on the control commands.

8. The adaptive adjustment method for a nano-grinding process according to claim 7, characterized in that, The double-nested optimization objective is decomposed into multiple subproblems based on the alternating direction multiplier method. The optimal solution of each subproblem is solved through alternating iterations, including: During the solution process, the inner optimization sub-problem satisfies the constraints that the trajectory coverage density is greater than or equal to the preset density threshold and the empty travel ratio is less than the preset ratio threshold. The outer optimization subproblem satisfies the constraint that the process entropy yield is less than the entropy yield threshold.

9. An adaptive adjustment system for a nano-grinding process, characterized in that, include: The acquisition module is used to acquire force signal data, acoustic emission signal data, temperature signal data and vibration signal data during the nano-grinding process, and combine them into grinding data, and acquire energy flow data. The extraction module is used to extract features from the grinding data, obtain feature parameters, and combine them with empirical wavelet transform to obtain the grinding state. The identification module identifies the grinding state based on topological data analysis and hidden Markov models, and obtains the identification results. The evaluation module obtains the process entropy yield based on energy flow data, performs performance evaluation based on the identification results and the process entropy yield, and obtains the evaluation results. The optimization module optimizes the process parameters based on the alternating direction multiplier method to obtain optimized process parameters; An adjustment module that adaptively adjusts based on optimized process parameters.

Images